# InfraRSI: Harnessing Recursive Self-Improvement for Training Infrastructure

DevFeed: [InfraRSI: Harnessing Recursive Self-Improvement for Training Infrastructure](<https://devfeed.tech/articles/infrarsi-harnessing-recursive-self-improvement-for-training-infrastructure-126138.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2611.11558>)

Author: InfraRSI Team, Yuchuan Tian, Haocheng Mei, Ye Yuan, Chao Xu, Boxun Li, Guohao Dai, Kai Han, Allen Chen, Yunhe Wang, Yu Wang

Published: 2026-11-10T05:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [AI Infrastructure](<https://devfeed.tech/topics/ai-infrastructure.md>), [distributed-training](<https://devfeed.tech/topics/distributed-training.md>), [LLM optimization](<https://devfeed.tech/topics/llm-optimization.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>), [Loop Engineering](<https://devfeed.tech/topics/loop-engineering.md>)

Tags: [agentic](<https://devfeed.tech/tags/agentic.md>), [agents](<https://devfeed.tech/tags/agents.md>), [compiler](<https://devfeed.tech/tags/compiler.md>), [costs](<https://devfeed.tech/tags/costs.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [distributed](<https://devfeed.tech/tags/distributed.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [experience](<https://devfeed.tech/tags/experience.md>), [foundation-model](<https://devfeed.tech/tags/foundation-model.md>), [gpus](<https://devfeed.tech/tags/gpus.md>), [infrastructure](<https://devfeed.tech/tags/infrastructure.md>), [kernel](<https://devfeed.tech/tags/kernel.md>), [launch](<https://devfeed.tech/tags/launch.md>), [llm](<https://devfeed.tech/tags/llm.md>), [llm-training](<https://devfeed.tech/tags/llm-training.md>), [measurement](<https://devfeed.tech/tags/measurement.md>), [memory](<https://devfeed.tech/tags/memory.md>), [mixture-of-experts-moe](<https://devfeed.tech/tags/mixture-of-experts-moe.md>), [models](<https://devfeed.tech/tags/models.md>), [recursive-self-improvement](<https://devfeed.tech/tags/recursive-self-improvement.md>), [training](<https://devfeed.tech/tags/training.md>)

## AI overview

InfraRSI treats distributed LLM training optimization as a coupled, verification-heavy problem: changes to launch settings, kernels, or communication can cause OOMs, compiler conflicts, noisy measurements, or numerical drift. Its five-stage workflow fixes workload, loss, and timing invariants; profiles baseline behavior and cluster noise; jointly searches launch settings and code patches; rechecks candidates; then audits and commits promoted changes. Instead of optimizing from one incumbent, it seeds diverse configurations from prior experience and models execution time and peak memory across trials, including OOMs. It uses those models to revise LLM estimates and combine memory-intensive changes with memory savers. Across four dense and MoE workloads on single and multiple nodes, the authors report up to 2.75x speedup and mean MFU rising from 17.4% to 30.0%, with experience transferred across GPUs.

## Source excerpt

arXiv:2611.11558v1 Announce Type: new Abstract: While Recursive Self-Improvement (RSI) accelerates isolated algorithm and kernel synthesis, distributed LLM training depends on tightly coupled launch settings, operator kernels, and communication schedules, where single edits often hit out-of-memory (OOM) boundaries, compiler conflicts, measurement jitter, or silent numerical drift. We present InfraRSI, the agentic harness that turns open-ended human commands into verified training optimizations via a five-stage workflow: init locks editable scope and a fixed task contract over workload, loss, and timing invariants to prevent reward hacking, doctor profiles baseline phases and calibrates cluster noise, run co-optimizes launch knobs and code patches before re-verifying top candidates against the baseline and logging findings to a cross-hardware experience bank, and report and apply audit source paths, track failed-trial costs, and commit promoted changes. Rather than stepping greedily from a single incumbent, InfraRSI initializes from diverse experience-seeded configurations and fits global execution-time and censored peak-memory models across all trials, including OOMs. Using these models, it prices per-setting memory costs, revises miscalibrated LLM priors from live runs, and pairs memory-hungry optimizations with memory savers and verified code patches. Across four single- and multi-node dense and Mixture-of-Experts (MoE) LLM workloads, InfraRSI achieves a speedup of up to $2.75\times$ and lifts mean Model FLOPs Utilization (MFU) from $17.4\%$ to $30.0\%$, outperforming existing tuners and coding agents while transferring experience across GPUs. As it improves training infrastructure efficiency, InfraRSI forms a two-level Recursive Self-Improvement loop---revising priors online within runs and distilling experience across clusters---to push foundation model scaling.